Detection Categories
How IdentiScope detects risk: all 17 harm categories, their keywords, and what to do
Harm
Highest priority. Requires immediate action.
Self-harm
Example keywords
kill myself
hurt myself
cutting
not worth living
end it all
unalive
What to do
Talk to your child today. Do not ignore this or wait. Approach calmly and without accusation. If you believe there is immediate risk, contact Samaritans (116 123, 24/7) or Pieta House. Consider speaking to a GP or counsellor.
Urgent Weight: 15 per hit
Suicidal ideation
Example keywords
suicide pact
pact to die
dying together
we made a deal
researching methods
What to do
This requires immediate attention. Do not wait. Call Samaritans on 116 123 (free, 24/7). In Ireland: Pieta House 116 123 or text HELLO to 50808. If there is immediate danger, call 999.
Urgent Weight: 20 per hit
Eating disorder
Example keywords
thinspo
proana
promia
meanspo
body check
restricting
calorie deficit
What to do
Approach calmly. Avoid commenting on weight or food. Ask how they are feeling about themselves. If concerned, speak to a GP or contact Beat Eating Disorders (beateatingdisorders.org.uk).
Urgent Weight: 13 per hit
Abuse
High priority. Speak to your child and monitor closely.
Bullying
Example keywords
kys
kill yourself
nobody likes you
expose
waste of space
everyone hates you
What to do
Ask if anything is happening with friends or at school. Check if they feel safe online. If this involves classmates, contact the school safeguarding lead.
Medium Weight: 4 per hit
Being bullied
Example keywords
being bullied
bullied at school
cyberbullied
spreading rumours
left out again
What to do
Ask gently how things are going with friends. Let them know they can come to you. Contact the school if classmates are involved. Consider Barnardos (barnardos.ie) for support.
Medium Weight: 7 per hit
Domestic abuse
Example keywords
scared to go home
afraid to go home
he touched me
not safe at home
smashed everything
What to do
Approach calmly and privately. Ask if they feel safe at home. Contact Women's Aid (Ireland: 1800 341 900) or the Gardaí if there is immediate danger.
Urgent Weight: 12 per hit
Image abuse
Example keywords
he shared my photos
posted my pictures
sent my nudes
get them taken down
non-consensual
What to do
Report to Gardaí immediately. Contact Hotline.ie to request image removal. This is a criminal offence under Coco's Law in Ireland. Support your child. It is never their fault.
High Weight: 12 per hit
Exploitation
High priority. Financial and substance exploitation of young people.
Drugs
Example keywords
cocaine
got a plug
weed
high af
gave me something
pingers
What to do
Talk about the risks of substance use without lecturing. Keep the conversation open. If you suspect regular use, speak to a GP or contact Drugs.ie (freephone 1800 459 459).
Medium Weight: 5 per hit
Muling
Example keywords
county lines
running packages
working the line
drug runner
easy money
trap line
What to do
Ask open questions about new friends or unexpected money. Children involved in county lines are victims, not offenders. Contact your local Garda station or Crimestoppers (1800 25 00 25).
High Weight: 10 per hit
Grooming & Sexual
High priority. Review contacts and report to CEOP if concerned.
Grooming
Example keywords
sneak out
home alone
parents gone
don't tell anyone
our secret
secret meeting
What to do
Review who they are talking to. Check plans for the coming days. Report concerns to CEOP (ceop.police.uk) or An Garda Síochána. Keep the conversation calm and non-accusatory.
High Weight: 10 per hit
Sexual content
Example keywords
loli
hentai
r34
sexting
nude
explicit
What to do
Have a calm, non-judgmental conversation about online safety, consent, and the risks of explicit content. If CSAM is involved, report to IWF (iwf.org.uk) or Hotline.ie immediately.
High Weight: 10 per hit
Sextortion
Example keywords
pay or i'll share
i'll post your nudes
send it to your school
threatening to share
i'm being blackmailed
What to do
Do not delete messages. They are evidence. Report to Gardaí or PSNI. Contact Hotline.ie or Revenge Porn Helpline (revengepornhelpline.org.uk). Remind your child they are the victim and this is never their fault.
High Weight: 14 per hit
Extremism & Ideology
High priority. Radicalisation indicators. Review accounts they follow.
Conspiracy
Example keywords
QAnon
WWG1WGA
5G causes cancer
microchips in vaccines
great reset
What to do
Talk about where they get information online and how to tell if something is reliable. Ask what makes them believe it. Check the accounts they follow for signs of radicalising communities.
Medium Weight: 4 per hit
Femcel
Example keywords
femcel
femcel community
looksmaxing
looksmaxxing
ugly and unwanted
What to do
Ask how they're feeling about themselves. These communities often push very negative ideas about appearance and self-worth. Check which online communities they're involved in. If you're concerned, the school guidance counsellor or deradicalise.ie can help.
High Weight: 12 per hit
Incel
Example keywords
incel
blackpilled
red pill
hypergamy
women only want chad
What to do
Talk about their views on relationships without lecturing. Look at who they're following and what communities they're in. A school counsellor or deradicalise.ie can help if you're concerned.
High Weight: 12 per hit
Far-right
Example keywords
white genocide
great replacement
white power
1488
based and redpilled
What to do
Look at who they're following and what communities they're in. Talk about what they believe, but be direct about the harms. Contact the school safeguarding lead or deradicalise.ie if you're concerned.
High Weight: 12 per hit
Far-left
Example keywords
smash the state
eat the rich
black bloc
burn it down
by any means necessary
What to do
Talk about what they believe. Political views are normal but there's a clear line between protest and content that pushes violence. Check what communities they're participating in.
High Weight: 12 per hit
Other Detection Signals
IdentiScope also looks at the account itself and how it's being used, not just what's being said.
Profile signals
- Bio contains predatory language (e.g. "DM me", "only fans", "18+" on a minor's account)
- Bio contains secrecy language (e.g. "don't follow me if you know me")
- No profile picture set
- Account created very recently (possible burner account)
- Very high follower ratio (many followers, few following, which is unusual for a teenager)
- Adult content setting enabled on account
Follower signals
- Following accounts with adult/explicit content in bio
- Following a high proportion of accounts with no profile picture
- Some followers use secrecy language or have hidden profiles
- Following a lot of significantly older accounts
Posting behaviour
- High proportion of posts between 1am and 5am (late-night usage)
- Sudden increase in post frequency (spike in activity)
- Predominantly negative sentiment across posts
- Very low engagement (likes/replies) relative to the number of posts, which may indicate blocked accounts or one-sided conversations
ML sentiment analysis
- Uses the Cardiff NLP sentiment model trained on 124 million tweets to classify each post as positive, neutral, or negative
- High negative sentiment across the full post history raises overall risk score
- Negative sentiment on flagged posts increases their individual weight
ML irony detection
- Uses the Cardiff NLP irony model to detect sarcasm and irony in posts
- Irony detection can downgrade a keyword hit, e.g. "oh great, another Monday, I want to die" said sarcastically is not treated as a crisis signal
- This reduces false positives on normal teenage venting
Evasion language
- Algospeak: substitute words used to bypass content moderation, e.g. "unalive" for suicide, "corn" for explicit content, "seggs" for sex
- Leetspeak: character substitution to bypass filters, e.g. "f4ck", "sh!t", "k!ll"
- Both categories flag that content is being deliberately obscured